145

*Author to whom all correspondence and reprint requests should be addressed:Sylvain Lehmann, CNRS, Institut de Génétique Humaine, Montpellier, France; CHU Montpellier, Labo-ratoire de Biochimie, Hôpital St. Eloi, Montpellier, France; Université Montpellier 1, F-34000, France. Email: Sylvain.Lehmann@igh.cnrs.fr.The first two authors contributed equally to this work.

Clinical Proteomics Copyright © 2006 Humana Press Inc.All rights of any nature whatsoever are reserved.ISSN 1542-6416/06/02:145–152/$30.00 (Online)

AbstractIn this manuscript, we compared serum pro-

files obtained with two related technologies,SELDI-TOF and Clinprot, using a single bioinfor-matic algorithm. These two approaches rely onmass spectrometry to detect proteins and pep-tides initially selected by binding to various chro-matographic matrices. They are proposed by two

different companies, and they are competing forbeing the reference in high throughput serumprofiling for clinical proteomics. This independentevaluation of these two technologies put the lighton some of their differences, suggests that theyaddress different proteome fractions and, thus,could be complementary. Taken together, ourdata could contribute to the parameters relevantfor the choice of one technology or the other.

Key Words: SELDI; Clinprot; profiling; serum;bioinformatics; clinical proteomics.

Introduction

Human serum and plasma have an importantclinical value for identification and detection

of biomarkers. However, the analysis of theseliquids is analytically challenging because of thehigh dynamic concentration range (over 10orders of magnitude) of blood constituentprotein/peptide species (1). High abundantproteins, such as albumin, immunoglobulins, or

Original Article

Comparison Between Surface and Bead-Based MALDI Profiling Technologies Using a Single Bioinformatics Algorithm

Christelle Reynès,1 Stéphane Roche,2,3 Laurent Tiers,2 Robert Sabatier,1 Patrick Jouin,4

Nicolas Molinari,5,6 and Sylvain Lehmann2,3,6,*1Laboratoire de Physique Moléculaire et Structurale, Faculté de Pharmacie, Montpellier, France; 2CNRS,Institut de Génétique Humaine, Montpellier, France; 3CHU Montpellier, Laboratoire de Biochimie,Hôpital St. Eloi, Montpellier, France; 4IGF, CNRS UMR5203, INSERM, U661, Univ Montpellier I,Univ Montpellier II, Montpellier, F-34094 France; 5Institut Universitaire de Recherche Clinique,Faculté Médecine, Montpellier; CHU Nîmes, Service DIM, Nîmes, France; and 6 Université Montpellier 1,F-34000, France

12_Lehmann 10/19/07 8:07 AM Page 145

Clinical Proteomics ________________________________________________________________ Volume 2, 2006

146 _______________________________________________________________________________ Reynès et al.

lipoproteins, produce large signals in most pro-teomic approaches and they mask or interferewith the detection of the other low amount pro-tein components. This situation explains why thediscovery of new protein or peptide biomarkersin blood is challenging. To minimize these prob-lems, separation proteomic scheme combiningfor example chromatography and mass spec-trometry (MS) methods were developed (2,3).This is the case of both surface-enhanced laserdesorption/ionization time-of-flight (SELDI-TOF) and ClinprotTM approaches (4), which relyon MS to detect proteins and peptides initiallyselected by binding to various chromatographicmatrices (anionic, cationic, IMAC, hydrophobic).These two approaches differ by the format ofthe chromatographic matrices, surface vsbeads, the mass spectrometers, and by the dataanalysis software used. They are proposed bytwo different companies, Ciphergen® (Fremont,CA) and Bruker Daltonics® (Bremen, Germany),respectively, and they are competing for beingthe reference in high throughput serum profil-ing for clinical proteomics. It is noteworthy thatresults obtained initially with this technologicalapproach have been often disappointing andcontroversial (5,6). However, other studies usingSELDI-TOF with protein identification and care-ful study design to avoid nonbiological arte-facts were able to demonstrate better outcomes,i.e., discovery and validation of potential cancerbiomarkers. An example is given by the mul-ticenter study by Zhang et al. (7) validatingthree biomarkers for the detection of earlystage ovarian cancer. Nevertheless, reductionof bias linked to preanalytical and analyticalphases, as well as use of prefractionationmethods (4,8–10), will most likely improve thepotency of these approaches in the future. Inthis work, we compared using a single bio-informatic algorithm, serum profiles obtainedwith SELDI-TOF and Clinprot. This indepen-dent evaluation of the relative performance ofthe two methods could help in choosing afuture serum profiling technology.

Experimental Procedures

Study Design and Biological Samples

To mimic a serum proteomic profilingexperiment run on the two technologies, weanalyzed a group of 12 serum samples fromC57BL/6 mice (collected between the age of150 and 250 d). Similar results were obtainedon human samples (not shown). Serum (100 µL)were obtained from 12 different mice by jugu-lar puncture as part of a control group for anongoing serum profiling experiment. Theblood was collected in Eppendorf tubes with-out additive, let clot 20 min at room tempera-ture and centrifuged for 20 min at 3000g.Serum was recovered and frozen at –80°Cuntil used.

SELDI-TOF Analysis

For SELDI-TOF analysis, each serum samplewas diluted 1.5 times with a solution of 8 Murea, 1% CHAPS, and shaken 15 min at roomtemperature. Denaturated samples were diluted40 times in the binding buffer (100 mMammonium acetate pH 4.0, 0.1% Triton) forapplication on CM10 (weak cation exchange)ProteinChip (Ciphergen). CM10 ProteinChiparrays were pre-equilibrated with 150 µL ofbinding buffer using a 96-well bioprocessorand incubated 5 min with gentle agitation.After removing the binding buffer from thewells, 100 µL of denaturated samples wereadded and incubated for 1 h on a plate shakerat room temperature. The wells were washedtwice with the binding buffer, once with 100 mM ammonium acetate pH 4.0 and finallyonce with water. ProteinChip arrays wereremoved from the bioprocessor and air-dried.Finally, 0.8 µL of α-cyano-4-hydroxycinnamic(CHCA) acid solution (10% in 50% acetonitrile,0.25% trifluoroacetc acid) was applied to eachspot and the chips were allowed to air-dryagain. Mass spectrometric analysis was per-formed by SELDI-TOF with a PBS-II Pro-teinChip reader (Ciphergen) using the same

12_Lehmann 10/19/07 8:07 AM Page 146

Volume 2, 2006 ________________________________________________________________ Clinical Proteomics

SELDI and Clinprot Comparison _____________________________________________________________ 147

settings for all the samples and for data collec-tion as follows: laser intensity 200, detectorsensitivity 7, molecular mass range 1000 to20,000 m/z, center mass 10,500 m/z, 160 shotsper spot. External calibration was done withthe All-in-1 Protein Standard II (Ciphergen).

ClinProt Analysis

Each serum sample was diluted 1.5 times ina solution of 8 M urea, 1% CHAPS, and shaken15 min at room temperature. Ten microliters ofMB-WCX (weak cation exchange) bindingsolution and 10 µL of MB-WCX beads wereadded to 5 µL of denatured samples. After a10-min incubation, microbeads were washedtwice using 100 µL of the MB-WCX wash solu-tion using the magnetic bead separator (MBS) tocollect the microbeads. After removal of thewash solution, 5 µL of MB-WCX elution solu-tion was added during 5 min. Microbeadswere then collected with the MBS; the super-natant was transferred into a fresh tube con-taining 5 µL of MB-WCX stabilization solution.Finally, 1 µL of the eluate was mixed 1:1 withthe CHCA solution (prepared as previouslydescribed) and 0.5 µL was applied on anAnchor chip sample plate. MS analysis wasperformed on an Ultraflex MALDI-TOF(Bruker Daltonics). The settings used were thefollowing: laser 20 ps (20 MHz), 25–35% power,sum up 1000 satisfactory shots in 100 shot steps,deflector set at 900 m/z and reflector off. Theuse of the MALDI-TOF in the linear mode,without reflector is adapted to the Clinprotapproach that necessitates detection of ionswith m/z values greater than 5000.

Exportation and Conversion of the Raw Data

SELDI spectra were exported as raw datausing the function provided in the Pro-teinChip software v3.2 (Ciphergen Biosys-tems). The generated file that contains theintensity values at all the m/z points wasimported in R using the function read.table().

R is a language and environment for statisticalcomputing and graphics (http://www.r-pro-ject.org/). R is available as Free Softwareunder the terms of the Free Software Founda-tion’s GNU General Public License in sourcecode form. The software used for this work isavailable upon request to C.R. For the Clinprotdata, the data are stored in a “fid” format thatwas converted into the “mzXML” formatusing the software Compass Xport 1.2.3(Brucker Daltonics). The data in the latterformat were imported in R thanks to thelibrary CaMassClass (11).

DATA Processing and Analysis

Combination of Clinprot Spectra

Bruker Daltonics recommended performingfour replicates per samples from the samemicrobeads separation probably as a mean toimprove the repeatability. Importantly, thefour replicates did not exactly have the samem/z coordinates, as a result of the mass spec-trometer variability, and therefore the simplemean between these spectra was not possible.The four spectra were therefore sorted byascending m/z and the average of 10 succes-sive points, belonging to the four spectra, wascalculated. This decreased the total number ofpoints par spectrum by a factor of two. How-ever, this point density was still higher thanthat of the SELDI-TOF spectra by a factor 1.2.

Detection of Peaks

The first step of this detection was repre-sented by the normalization of the spectra. Todo so, the mean intensity in the range 1500 to10,000 m/z was calculated for each individualspectra and for each technology. A normal-ization coefficient was defined for each spec-trum as the ratio global/individual mean and applied. This normalization method isstandard and is used in particular in theCiphergen software. Peak detection was thenperformed for each spectrum using the

12_Lehmann 10/19/07 8:07 AM Page 147

Clinical Proteomics ________________________________________________________________ Volume 2, 2006

148 _______________________________________________________________________________ Reynès et al.

following method: first, the spectrum wasdivided into two equal parts. In each part, theintensity maximum was identified. Then theboundaries of the corresponding peak werelocated based on the sign changes of the firstderivative of the spectrum. For derivativecomputation, the spectrum was temporarilysmoothed using Friedman’s super smoother(12). These boundaries became the new limitsof new zones in which a new local maximumwas looked for. This sequence was repeatedstops until the distance between two bound-aries was smaller than the mass accuracy (i.e.,0.1% as provided by the companies and veri-fied on the spectra). Then, based on the distri-bution of the valley-depths of all the peaksfound in all the spectra (for each technology),a threshold was chosen, below which thepeaks can be considered as noise. This thresh-old was determined graphically by locatingthe intensity below which frequency of pointsis abnormally high (results not shown).

Alignment of the Spectra

To compare the data generated within eachtechnology and determine if peaks present indifferent spectra arisen from the same pep-tide/protein species, an alignment was real-ized as follows. The m/z locations of all thepeaks from all the spectra were collected andsorted in ascending order. Then, a hierarchicalclustering approach was applied to obtainpeak clusters which minimum size corre-sponded to the mass accuracy value.

Comparison of Peaks Between the Two Technologies

Once the peaks were selected for bothSELDI-TOF and Clinprot (see Detection ofPeaks), the clustering was performed betweenthe two technologies using the peaks identifiedfollowing the alignment of all the spectra. Thesame clustering method was used, but as onecould consider that there is a shift between the

two technologies, the threshold used corre-sponded to twice the mass accuracy.

Results and Discussion

The purpose of this work was to compareproteomic profiles obtained with two relatedapproaches, SELDI-TOF and Clinprot. Thesetwo leading profiling technologies are pro-posed by two different companies, Ciphergenand Bruker Dynamics, respectively. We carryout this study using as initial step of the pro-filing, the capture of proteins on comparableweak cation exchange chromatographic matri-ces, coupled to surfaces (CM10, SELDI-TOF)or microbeads (WCX, Clinprot). Twelve mousesera were analysed using recommended ana-lytical protocols and the same CHCA matrixfor MS. The idea was to mimic a small groupof serum samples, as analyzed in many serumprofiling studies (13). To avoid bias related tothe different software used by the two compa-nies, raw data were exported to the statisticalsoftware R before normalization and peakdetection (see Experimental Procedures). Wefocused our analysis on the 1500 to 10,000m/z range which is optimal with the CHCAMS matrix used.

A first difference between the two types ofspectra lied in the density of points generated.In fact, between 1500 and 10,000 m/z the Clin-prot spectra were constituted of 106,431 +/–4089 points, whereas the SELDI-TOF had only48,410 points. These values are chosen by thecompanies, and are linked to the performancesof the two mass spectrometers used. This dif-ference in density was partially accountable fordifferences in background signal variability, or noise (Fig. 1A). In fact the noise wassignificantly lower in Clinprot, than in SELDI-TOF, as confirmed by its variance of 1461.93and 6008.22, respectively (p < 0.0001, F-test).The scale of intensities of the spectra was alsodifferent in the two technologies as the rangefor the Clinprot data went from 0.70 to

12_Lehmann 10/19/07 8:07 AM Page 148

Fig.

1.(A

)D

istr

ibut

ion

of t

he b

ackg

roun

d si

gnal

var

iabi

lity,

or n

oise

,for

Clin

prot

(gr

een)

and

SEL

DI-T

OF

(red

) be

twee

n 64

00 a

nd 6

700

m/z

.N

ote

that

the

dis

trib

utio

n is

mor

e ho

mog

eneo

us i

n C

linpr

ot.(

B)

Gen

eral

vie

w o

f a

repr

esen

tativ

e SE

LDI-T

OF

(red

) an

d C

linpr

ot (

gree

n)se

rum

pro

file.

In t

he in

sert

,som

e pe

aks

are

pres

ent

in o

ne t

echn

olog

y an

d no

t in

the

oth

er (

star

s),w

here

as o

ther

s (t

rian

gle)

are

com

mon

to b

oth.

(C)

Com

pari

son

for

the

two

tech

nolo

gies

of

the

reso

lutio

n ob

tain

ed f

or t

he s

ame

peak

loc

ated

nea

r 28

00 m

/z.(

D)

Gen

eral

over

view

of t

he 1

2 sp

ectr

a fr

om t

he t

wo

tech

nolo

gies

bet

wee

n 15

00 a

nd 1

0,00

0 m

/z.T

he p

rese

nce

of a

dditi

onal

pea

ks in

the

hig

h m

/z r

atio

is c

lear

ly v

isib

le in

SEL

DI-T

OF.

149

12_Lehmann 10/19/07 8:07 AM Page 149

Clinical Proteomics ________________________________________________________________ Volume 2, 2006

150 _______________________________________________________________________________ Reynès et al.

81828.43 and for SELDI-TOF from –11.28 to342.56. To facilitate the comparative analysis ofthe two types of spectra, the values of the rawSELDI intensities were multiplied by 1000 andused as a common arbitrary unit for the inten-sity. This did not affect the overall analysis ofthe spectra as the same peaks were detectedbefore and after applying the multiplicationfactor (not shown). Importantly, for the analy-sis of the Clinprot data, Bruker Daltonics rec-ommended performing four replicates of eachspectrum from the same microbeads separa-tion. To conform to this recommendation, themean of these four spectra was calculatedbefore analysis (see Experimental Procedures).

As illustrated Fig. 1B, the general aspects ofSELDI-TOF and Clinprot spectra obtainedusing similar capture matrices were alike.However, differences in terms of peak presenceor absence, height, and resolution were clearlyapparent (insert, Fig. 1B). The latter parameteris important for the detection and quantifica-tion of different peaks; a high resolution leadsto rapid comeback to the baseline and a goodseparation of two peaks without contamina-tion of each species. In our case, the CiphergenPBSIIc mass spectrometer has a lower resolu-tion, as illustrated in the vicinity of the 2800 m/z peak (Fig. 1C). This difference withthe Clinprot Ultraflex I mass spectrometer willbe reduced with the new generation of Cipher-gen mass spectrometer (PBS4000). Interestingly,the difference in resolution did not dramatically

modify the total number of peaks detected inboth technologies (see Table 1).

To validate our observation independentlyfrom a particular sample, we have performedthe analysis of 12 different mouse sera usingboth technologies (Fig. 1D). The detection of thepeaks in all the spectra was realized based onsign changes of the derivated spectra. An equiv-alent number of peaks (close to 80 between 1500and 10,000 m/z, see Table 1) was detected inboth technologies. Interestingly, analysis of the SELDI-TOF spectra with the Ciphergenbiomarker software also resulted in an average80 peaks detected when a signal/noise ratio of three was used (not shown). This validates the performance of our biostatistical method.Importantly, significant differences were observedbetween the two technologies for the peak dis-tribution in regards to the m/z values (Fig. 2A).In fact in the lower m/z range, less than 5000m/z, Clinprot could detect more peaks thanSELDI-TOF, whereas above this value, it was theopposite (Table 1). This difference is most likelyrelated to the higher resolution of the Clinprotmass spectrometer that resolves more peaks forsmall peptides. A high MS resolution is in factessential for peptide mass fingerprint and iden-tification purposes (14). It is also valuable forprofiling of small ions, but based on our resultsit seems less critical here (in the high massrange) because we analyzed nonproteasedigested proteins from complex biological sam-ples like serum.

Table 1Summary of the Peaks Detected for the 12 Samples, in Both Technologies,

and Their Relative Distribution Between Low and High m/z Ratio

Statistical analysisSELDI-TOF Clinprot (t student)

Total number of peaks detected 80.8 ± 27.9 80.3 ± 13.6 p = 0.95621500 < m/z < 10,000

Peaks between 55.2 ± 17.7 75.0 ± 12.3 p < 0.0051500 < m/z < 5000 (83% ± 5%) (96% ± 1%)

Peaks between 10.66 ± 3.42 3.2 ± 1.03 p < 0.0015000 < m/z < 10,000 (17% ± 5%) (4% ± 1%)

12_Lehmann 10/19/07 8:07 AM Page 150

Volume 2, 2006 ________________________________________________________________ Clinical Proteomics

SELDI and Clinprot Comparison _____________________________________________________________ 151

To directly compare the results obtained inthe two technologies, the alignment of thepeaks between all the spectra was realizedusing hierarchical clustering with a thresholdcorresponding to the m/z accuracy. Twentyfive m/z peaks were detected in more thanhalf the spectra in the two technologies. Theintensity of these twenty five peaks wascorrelated between the two technologies (seeexample Fig. 2B, correlation factor = 0.84 ± 0.1).

This suggested that the binding and the detec-tion of common peaks were somehow compa-rable in the two technologies. However, as wementioned before, many peaks were detectedonly in one technology or the other, as illus-trated by the result of the hierarchical cluster-ing realized between SELDI-TOF and Clinprotpeaks (Fig. 2C).

Taken together our results indicate thatSELDI-TOF and Clinprot technologies could

Fig. 2. (A) Histogram of the m/z distribution for the peaks detected in all the spectra in the two technologies.Clinprot identifies more peaks in the lower m/z whereas SELDI-TOF shows more peaks with high m/z.(B) Comparison of the profiles obtained by the two technologies on the same sample for 25 common peaks:the intensities are correlated despite a bigger variance for peaks with high intensities. (C) Representation ofthe presence (in red)/absence (in black) of all the detected peaks in each spectrum for the two technologies.On the left, stand the peaks common to both technologies and on the right those specific to one of them.

12_Lehmann 10/19/07 8:07 AM Page 151

Clinical Proteomics ________________________________________________________________ Volume 2, 2006

152 _______________________________________________________________________________ Reynès et al.

achieve a comparable proteomic profiling fromunfractionated serum which could then beused for detection of potential blood biomark-ers. However, the ClinProt technology allows toanalyse, for one sample, not only the subset ofproteins retained by the chromatographic sur-face as in SELDI-TOF, but also the nonretainedfraction and the eluted fractions, as on chro-matographic columns. This represents an attrac-tive possibility for this technology, which alsoallows the use of several type of MS matricesfor a single capture experiment. The use of amass spectrometer with a better resolution, hereUltraflex I vs PBSIIc, and for SELDI the newPBS4000 vs PBSIIc, facilitates peaks detectionand quantitation (especially in the lower m/zrange) and should be favoured. Interestingly,although some peaks appeared to be present inboth profiles using the two technologies, manydifferences in the profiles still exist suggestingthat they address different proteome fractionsand could be complementary. In conclusion, ourstudy does not definitely favor the choice of onetechnology or the other, and additional param-eters like purification procedures of candidates,cost, or possibilities for clinical multisite valida-tion, need to be taken into account before choos-ing between these two approaches.

Acknowledgments

We thank Prof. Jean-Paul Cristol for his sup-port. Supported by grants from the CHUMontpellier AOI 2004, the EC network ofExcellence “Neuroprion” FOOD-CT-2004-506579and the CNRS.

References1. Anderson, N. L. and Anderson, N. G. (2002)

The human plasma proteome: history, charac-ter, and diagnostic prospects. Mol. Cell. Pro-teomics 1, 845–867.

2. Pieper, R., Gatlin, C. L., Makusky, A. J., et al.(2003) The human serum proteome: display ofnearly 3700 chromatographically separatedprotein spots on two-dimensional electro-

phoresis gels and identification of 325 distinctproteins. Proteomics 3, 1345–1364.

3. Yang, Z., Hancock, W. S., Chew, T. R., andBonilla, L. (2005) A study of glycoproteins inhuman serum and plasma reference standards(HUPO) using multilectin affinity chromato-graphy coupled with RPLC-MS/MS. Pro-teomics 5, 3353–3366.

4. Hortin, G. L. (2006) The MALDI-TOF massspectrometric view of the plasma proteomeand peptidome. Clin. Chem. 52, 1223–1237.

5. Petricoin, E. F., Ardekani, A. M., Hitt, B. A., et al. (2002) Use of proteomic patterns in serumto identify ovarian cancer. Lancet 359, 572–577.

6. Coombes, K. R., Morris, J. S., Hu, J., Edmon-son, S. R., and Baggerly, K. A. (2005) Serumproteomics profiling: a young technologybegins to mature. Nat. Biotechnol. 23, 291–292.

7. Zhang, Z., Bast, R. C., Jr., Yu, Y., et al. (2004)Three biomarkers identified from serum pro-teomic analysis for the detection of early stageovarian cancer. Cancer Res. 64, 5882–5890.

8. Issaq, H. J., Conrads, T. P., Janini, G. M., andVeenstra, T. D. (2002) Methods for fractiona-tion, separation and profiling of proteins andpeptides. Electrophoresis 23, 3048–3061.

9. Pieper, R., Su, Q., Gatlin, C. L., Huang, S. T.,Anderson, N. L., and Steiner, S. (2003) Multi-component immunoaffinity subtraction chro-matography: an innovative step towards acomprehensive survey of the human plasmaproteome. Proteomics 3, 422–432.

10. Guerrier, L., Thulasiraman, V., Castagna, A., et al. (2006) Reducing protein concentrationrange of biological samples using solid-phaseligand libraries. J. Chromatogr. B Analyt. Technol.Biomed. Life Sci. 833, 33–40.

11. Tuszynski, J. (2006) caMassClass: processingand classification of protein mass spectra(SELDI) data. http://cranr-projectorg/src/contrib/Descriptions/caMassClasshtml (ThecaMassClass Software License, Version 1.0).

12. Friedman, J. (1984) A variable span scatterplotsmoother: Laboratory for ComputationalStatistics, Stanford University, report no.: Tech-nical Report No. 5.

13. Fung, E. T., Wright, G. L., Jr., and Dalmasso,E. A. (2000) Proteomic strategies for biomarkeridentification: progress and challenges. Curr.Opin. Mol. Ther. 2, 643–650.

14. Cottrell, J. S. (1994) Protein identification by pep-tide mass fingerprinting. Pept. Res. 7, 115–124.

12_Lehmann 10/19/07 8:07 AM Page 152